L.R. Carley

Performance comparison of detection methods in magnetic recording

Various detection schemes suitable for magnetic recording are compared in terms of their effective signal-to-noise ratios. It is shown that at high densities the performance of conventional detectors such as a peak detector, a threshold detector with partial response equalization, a decision feedback equalizer, and a Viterbi algorithm detector tuned to a linearly truncated channel fall far below the optimum performance that can be achieved by the maximum-likelihood sequence detector (MLSD). It is shown that while implementing the full MLSD is clearly out of the question for high densities with severe intersymbol interference (ISI), there exists an efficient detection scheme which achieves an excellent compromise between hardware complexity and detection performance. This scheme, which is called the fixed-delay tree search with decision feedback, combines a fast and efficient tree-search algorithm with a decision feedback equalizer to cancel out a portion of ISI without noise enhancement. It is well suited for run-length-limited systems and attains performance close to that of MLSD while maintaining reasonable implementation cost and processing requirements. >

Efficient sequence detection for intersymbol interference channels with run-length constraints

This paper addresses the data detection problem of intersymbol interference (ISI) channels with a specific modulation code-constraint known as the (d, k) run-length-limited (RLL) constraint, a popular modulation code-constraint for data storage channels as well as certain communication channels. A computationally efficient sequence detection algorithm is proposed which yields a performance close to that of the maximum likelihood sequence detector when applied to such ISI channels. The proposed detector is derived as a high signal-to-noise ratio approximation to the delay-constrained optimum detector, one which minimizes the symbol error probability given a fixed decision-delay constraint. The proposed algorithm is essentially a fixed-delay tree search (FDTS) algorithm with systematic ambiguity checking and is closely related to existing finite-depth tree search algorithms. It is observed that long critical error events common in uncoded ISI channels are eliminated by the RLL constraint. Based on this observation, we show that for some important RLL constrained channels, the proposed FDTS algorithm yields the same minimum Euclidean distance between distinguishable channel output sequences as the unconstrained maximum likelihood sequence detector. >

System design considerations for MEMS-actuated magnetic-probe-based mass storage

This paper presents common system design considerations imposed on magnetic storage devices that employ MEMS devices for positioning of a magnetic probe device over a magnetic media. The paper demonstrates that active servo control of the probe tip to media separation can be achieved with sub-nanometer accuracy. It demonstrates that reasonable-size capacitive sensors can resolve probe tip motions with a noise floor of roughly 22 picometers, allowing them to be used as position sensors in magnetic force microscope (MFM) readout approaches. In addition, this paper demonstrates that although MEMS media actuators can achieve scanning ranges of /spl plusmn/50 um, the mass of the media sled imposes important access time and data rate constraints on such MEMS-actuated mass storage devices.

Comparison of computationally efficient forms of FDTS/DF against PR4-ML

The authors compare the error-rate performance and circuit complexity of fixed-delay tree search with decision-feedback (FDTS/DF) to that of Class IV partial response equalization followed by Viterbi detection (PR4-ML). The impact of realistic impulse responses, misequalization, phase errors, and offtrack interference on both detection strategies is explored. Two computationally efficient methods for implementing FDTS/DF are presented. These FDTS/DF architectures avoid explicit multiplication-one uses a RAM-based table lookup and the other uses binary equalization over the tree-depth. >

Detection performance in the presence of transition noise

The performance of various detector/RLL (run-length-limited) code combinations was investigated assuming the presence of both additive white noise and transition noise. The results indicate that for detectors relying heavily on linear suppression of ISI (intersymbol interference), the transition-noise effect does not show up at high densities because of relatively severe enhancement of the additive-noise component. However, transition noise degrades performance of FDTS/DF (fixed delay tree search with decision feedback) and DFE (design feedback equalizer), as a large amount of ISI is allowed at the equalizer output to avoid the excessive additive-noise enhancement. Nevertheless, the FDTS/DF still has substantial performance gain over other detectors. Effects of position jitter and width variations have also been observed separately. It was found that the simple peak detector performs comparatively well in the presence of position jitter, whereas sequence detection schemes that permit a relatively large amount of ISI at the linear equalizer output are more sensitive to position jitter than to width variations, except at low uncoded-densities. It was also observed that the signal-to-noise ratio penalty incurred by the d-constraint is significantly reduced in the transition-noise environment compared to the additive-noise environment. >

Equalization and detection for nonlinear recording channels with correlated noise

At high recording densities, signal nonlinearity due to both the medium and the magnetoresistive head and correlated noise due to the medium and the equalization process, substantially degrade the error-rate performance of the Viterbi algorithm (VA). This paper describes a method for dealing with both problems by simultaneously adapting the equalizer, its target and the VA branch metrics. A monic constraint is employed to locate a target that limits noise correlation at the input to the VA. Pattern-dependent offsets, rather than actual VA ideal values, are adaptively determined to account for channel nonlinearity in the VA. The effectiveness of these techniques is demonstrated on very high density data collected from a spin stand.

Adaptive continuous-time equalization and FDTS/DF sequence detection

An adaptive analog continuous-time forward equalizer and a Fixed Delay Tree Search with Decision Feedback (FDTS/DF) detector can be used to implement high speed low power sequence detection for a magnetic recording system. An analog continuous-time forward equalizer offers a low power, high speed solution to the equalization problem, although it cannot achieve the degree of pulse shape control provided by a long FIR filter. However, an FDTS/DF detector can be easily adapted to cope with the resulting imprecisely equalized pulses. Earlier work demonstrated the suitability of a Complex Graphic Equalizer (CGE) for the adaptive continuous time forward equalizer. System simulations indicated the viability of a CGE-FDTS/DF detector in a 2/3(1,7) coded channel. New results explore the performance of a CGE-FDTS/DF system for an 8/9(0,3) coded channel. Additionally, the feasibility of a CGE-EPR4 detection system is demonstrated.

Adaptive continuous-time equalization followed by FDTS/DF sequence detection

Employing an adaptive continuous-time forward equalizer with a Fixed Delay Tree Search with Decision Feedback (FDTS/DF) detector provides several advantages to a magnetic recording system. Although an adaptive continuous-time filter of modest complexity cannot achieve the degree of pulse shape control provided by a long FIR filter, its power consumption may be dramatically lower. An FDTS/DP detector can be easily adapted to widely varying equalized pulses. By combining the two, we can achieve channel performance comparable to that of a system including an FDTS/DF detector and a long FIR equalizer-but with a power dissipation that is potentially much less than conventional sequence detection channels. Simulation results are presented that compare this new approach with a finite length FIR equalizer followed by an FDTS/DP detector. We present and compare several architectures for the adaptive continuous-time equalizer that offer improved convergence when using the least-mean-squares (LMS) adaptation algorithm. >

A pipelined 16-state generalized Viterbi detector

An architecture for efficiently implementing linear and nonlinear generalized Viterbi detectors is presented. Its principle novelty is the use of a pipelined look-up table that stores the squared difference between every possible input level and every noiseless level. An implementation example is presented for the case of a 16-state Viterbi detector handling any noiseless target of up to 5 adjacent non-zero values. In addition, auxiliary circuits to generate the error between the input and the noiseless target value (needed to implement adaption of noiseless target values and equalizer coefficients, timing recovery, automatic gain control and DC offset cancellation) are presented.

Time measurements of amplitude and bit shift in thin metallic media

Improved models for media noise are a prerequisite for the design of recording Channels that utilize an increased fraction of the information capacity (linear bit density) available to a given head/disk interface. Noise measurement techniques using power spectral density or RMS noise power inherently assume that all noise sources are stationary and, thus, cannot accurately predict the time-varying statistics of a non-stationary noise process. Studies have indicated that noise in metallic thin-film media is localized in the transition regions; hence, it is not a stationary process. Therefore, we have chosen to characterize media noise by direct time measurement of the readback voltage. In this paper we model thin metallic media noise by assuming that the magnetic transitions experience random variations in both position and width. A time measurement technique was employed to estimate the second order statistics of width and position of isolated readback pulses in 3 different samples of metallic thin-film media. The RMS noise estimated from the model and time measurements of the sample statistics accurately predicts the RMS noise measured by a spectrum analyzer, supporting the validity of the proposed model.